Fill-Side Behavioral Concentration on Polymarket: Identification Limits under Record-Level Attribution

Battle Count: 55

The paper is highly relevant for prediction market trading, leveraged event-linked perpetual futures design, and risk calibration. It establishes critical identification limits: public fills cannot reveal quote lifecycles, market-making roles, or manipulation intent. The concentration findings (12.6% of addresses holding 81.4% of attributed notional) are directly relevant for liquidity risk modeling and adverse selection estimation. However, the paper is primarily a methodological correction and identification-limit paper rather than a trading strategy paper. Its value to quantitative trading lies in constraining what can be inferred from on-chain data, informing risk engine parameterization, and guiding surveillance system design. The match-fragmentation non-invariance result is critical for anyone building features from CLOB fill logs.

The conceptual framework (propositions on non-invariance, non-identification, and one-cluster interpretation) is mathematically straightforward but requires careful reading. The empirical pipeline involves processing 13.36M on-chain records, constructing six-dimensional feature vectors, running DBSCAN across 15 configurations, and applying threshold cohort rules. The required replication (match normalization, mint/burn-aware direction, multi-window analysis) would be substantially more complex. The paper itself performs no new computation, so implementation of the paper's content is minimal; implementing the proposed replication design is moderately to highly complex.

The paper explicitly states it performs no new empirical run and retains original numerical outputs only as results under the legacy-CTF, record-level two-sided attribution convention. No code or data release is provided. The data and code statement calls for a future public release with exact block/contract registries, match-construction rules, execution-type classification, and reconstruction procedures. The archived extraction parameters (block range, contract filter, attribution convention) are stated, enabling partial reproduction from public Polygon data, but the exact processing pipeline is not released.